This invention relates to the inspection of flat patterned media using optical techniques. More specifically, this invention relates to the automated optical inspection (AOI) of large flat patterned media, such as thin film transistor (TFT) arrays (the main component of liquid crystal flat panel displays (LCD)). Although the invention is applicable to the general case of inspection of any flat, periodically patterned media, it is particularly useful for the high throughput, in-line inspection of TFT arrays at various stages of their production.
During the manufacturing of LCD panels, large clear sheets of thin glass are used as a substrate for the deposition of various layers of materials to form electronic circuits that are intended to function as a plurality of separable, identical display panels. This deposition is usually done in stages, where in each stage, a particular material (such as metal, Indium Tin Oxide (ITO), Silicon, Amorphous Silicon, etc.) is deposited over a previous layer (or upon the bare glass substrate) in adherence to a predetermined pattern. Each stage includes various steps such as deposition, masking, etching, and stripping.
During each of these stages and at various steps within a stage, many production defects may occur, that may have electronic and/or visual implications for the performance of the final LCD product. Such defects include, but are not limited to: circuit shorts, opens, foreign particles, miss-deposition, feature size problems, and over and under etching. The most common defects, shown in FIG. 1, include: metal protrusion 110 into ITO 112, ITO protrusion 114 into metal 116, a so-called mouse bite 118, an open circuit 120, a short 122 in a transistor 124, and a foreign particle 126.
In a specific application domain, such as TFT LCD panel inspection, the article defects subject to detection are small (down to individual micrometers), thus requiring demanding defect detection limits.
However, mere detection of defects is insufficient. Detected defects must also be classified as process defects, i.e. minor imperfections, which do not undermine the performance of the finished product but are an early indication of the array manufacturing process drifting out of optimum conditions; reparable defects, which can be repaired to improve the array production yield; and finally killer defects, which disqualify the TFT array from further use.
In any conventional AOI system, there is always a compromise between a number of critical characteristics, such as the optical scanning resolution, tact time, detection limits, and cost. These characteristics determine the usefulness or type of application of the AOI instrument. Typically, one characteristic can be optimized or improved by compromising another. For example, the AOI system resolution can be increased, resulting in improved detection limits (smaller defects detectable); these improvements would, however, have an adverse effect on the time needed to complete the inspection (tact time) or the system cost. Conversely, for a different type of application, the detection limits can be relaxed (larger defects detectable) by lowering the system resolution, therefore achieving a shorter tact time and reduced system cost.
The inability of the present art to provide high detection sensitivity and tact time matched to the production speed at an acceptable price has imposed on the LCD industry the use of low performance, short tact time systems as in-line instruments. Higher detection sensitivity systems (always requiring long inspection times and incompatible with the production speed) could only be used as off-line instruments, capable of inspecting only selected TFT panels. This method of inspection is often referred to as the sampling mode of operation.
The operating resolution of an AOI system has a direct impact on its cost. For a short tact time, this cost increases almost exponentially with the increase in operating resolution. Therefore, for high-throughput, in-line applications at production speeds, where a short tact time is required, only comparatively lower resolutions have been feasible for the system.
For the application domain of interest, conventional AOI systems use variants of the pattern comparison technique used to detect the presence and location of defects. These methods take advantage of the periodical character of the object under test and directly compare regions spaced by the pattern period or its multiples.
Those skilled in the art may recognize that prior spatial domain image comparison techniques suffer from the pixelation effect, which always degrades the detection limits of the system. The pixelation effect, often interpreted as noise within an image, becomes especially significant in the vicinity of circuit features where rapid transitions of image intensity occur. This leads to false detections or masking of legitimate defects. These effects are highly undesirable since the inspection instruments are expected to have the highest detection sensitivity in the vicinity of TFT array features, such as transistors, data and gate line crossings. Pixelation suppression methods, such as those primarily based on sub-pixel interpolation and approximation techniques, are used as means to partially alleviate these shortcomings. These methods nevertheless fail to satisfy the demands of this particular application domain.
These inherent limitations have led developers to explore the promise of suppressing observable periodic patterns in the optical domain before pixelation is introduced by digitization of the object image for analysis. It is known, for example, that a well understood property of lenses is the ability to form a two dimensional Fourier spectrum of the object in the lens focal plane. The Fourier transformation occurs entirely in the optical domain before any digitization of the signal. This presents the opportunity to filter periodic patterns in the image spectrum in an optical, analog manner.
Optical filtering presupposes the availability of a suitable spatial light modulator (SLM) placed in the lens focal plane to selectively attenuate the intensity profile formed therein to yield an altered (filtered) image in the image plane. The final image would ideally be digitized through an image capture device, such as a Charge Coupled Device (CCD) sensor. In the particular case of a periodically patterned surface as the object to be inspected, the ideal focal plane intensity profile is a well defined grid of intensity peaks. By means of an ideal specialized filter on the focal plane to mask out these peaks, it was expected that the periodic components of the image would be attenuated while preserving non-periodic signal components, such as those caused by defects in the original pattern.
However, the success of this filtering is highly dependent on a number of the properties of the mask placed in the focal plane, including contrast ratio, spatial resolution, optical quality, and the ability to be reprogrammed with high speed. The lack of suitable technology to implement an SLM with the required properties has rendered the optical Fourier filtering principle impractical in the application domain of interest.
AOI equipment has been characterized by a variety of problems. Most of the implemented solutions are based on spatial domain pattern comparison techniques often used in combination with sensor-level pixel or sub-pixel precision alignment techniques.
U.S. Pat. No. 4,579,455 to Levy et al. describes an alignment and pattern comparison technique where a pair of 7×7 windows are considered on the test and reference images and a squared sum of errors over a multitude of possible 3×3 sub-windows within this window are computed. If the minimum error over these twenty-five combinations exceeds a threshold value, a defect is assumed. The method appears to be capable of compensating for alignment mismatch down to a sensor pixel level.
Arguing about the coarse alignment precision of the method by Levy et al., U.S. Pat. No. 4,805,123 to Specht et al. describes an improved alignment and comparison technique for the detection of defects. In this technique, large windows in test and reference images are used to compute a sensor pixel level correlation between a test image and a reference image. The resulting sampled correlation surface's minimum point is found and a quadratic function is fit to the surface in the neighborhood of this minimum point. Using the fitted quadratic function, a sub-pixel precision translation is obtained to align the test and reference images. The aligned images are compared by thresholding image differences on 2×2 sub-windows on the test and aligned reference images.
Variations and improvements on these basic techniques have also been proposed. For example, U.S. Pat. No. 5,907,628 to Yolles et al., among other things, points out the drawbacks of using the sampled correlation surface to find the minimum and argues that due to a coarse sampling of the surface this point may not correspond to the true minimum. Hence, they argue that the subsequent sub-pixel interpolation step would do little to improve the detected minimum and a false alignment would result, leading to false alarms in detection. Yolles et al. proposes to alleviate these problems by an elaborate comparison process based on improved comparison entities.
Although pattern comparison based instruments have been successfully used in the industry for certain applications, the compromise between the system speed (tact time) and accuracy (defect detection limits) has been a dominant factor imposing what has been considered fundamental limitations. This ever present compromise for conventional inspection methods limits the usefulness of these systems in the application domain of high-throughput, in-line inspection of large flat patterned media at production speeds, where there is concurrent demand for speed and detection sensitivity.
The so called Optical Fourier Filtering (OFF) (also known as Fourier Spatial Filtering) is a known and understood technique. The attempts to use OFF for repetitive patterned surface defect detection dates back to a paper by Watkins entitled “Inspection of integrated circuit photo masks with intensity spatial filters,” published in Proceedings of the IEEE vol. 57, No 9, (September 1969), wherein the properties of such spatial filters are described. This was followed by the invention reported in U.S. Pat. No. 4,000,949 to Watkins et al describing fundamental aspects of basic OFF for patterned surface inspection.
As the technology to implement Fourier spatial filters proved feasible in certain application areas, a number of contributions followed. These include, for example, U.S. Pat. No. 4,806,774 to Lin et al., where a basic bright field illumination Fourier spatial filtering setup is described for microcircuit die inspection, and U.S. Pat. No. 5,383,056 and No. 5,627,678 both to Nishii et al., where particular lens arrangements with favorable properties are described. In U.S. Pat. No. 5,276,498 to Galbraith et al., another application of the Fourier spatial filter to highly periodic semiconductor wafer inspection is presented. The described system is designed for scanning a surface incorporating two regions with different light diffracting patterns. This surface is scanned by a narrow beam of light. The system implements a programmable SLM through the use of two successive stages of light valves, each stage being composed of a one-dimensional array of light valves forming linear stripes and being in transverse configuration to one another.
Another proposed system in U.S. Pat. No. 5,506,676 to Hendler et al considers a spatial separator, such as a micro-mirror device, to redirect different parts of the lens focal plane information into different light intensity sensors for parallel analysis and The system does not incorporate an image capture device.
Other U.S. patents which disclose relevant art include:                U.S. Pat. No. 6,490,393 to Zhou, entitled Integrated optical multiplexer and demultiplexer for wavelength division transmission of information;        U.S. Pat. No. 6,137,570 to Chuang et al. entitled System and method for analyzing topological features on a surface;        U.S. Pat. No. 6,128,078 to Fateley entitled Radiation filter, spectrometer and imager using a micro-mirror array;        U.S. Pat. No. 6,084,671 to Holcomb entitled Surface analysis using Gaussian beam profiles;        U.S. Pat. No. 6,061,126 to Yoshimura et al. entitled Detecting system for surface form of object;        U.S. Pat. No. 6,046,808 to Fateley entitled Radiation filter, spectrometer and imager using a micro-mirror array;        U.S. Pat. No. 5,966,212 to Hendler et al. entitled High-speed, high-resolution, large area inspection using multiple optical Fourier transform cells; and        U.S. Pat. No. 5,822,055 to Tsai et al. entitled Optical inspection of a specimen using multi-channel responses from the specimen using bright and darkfield detection.        
Despite their inherent limitations, which cause high false alarm rates or limited sensitivity through relaxed thresholds, conventional pattern matching techniques have remained the dominant inspection techniques throughout the industry. This appears to be due to the nature of the sampling process, which does not attempt to perform 100% inspection at production speeds and does not emphasize the missed defect rate for the inspection system. However, the emerging application domain of high-throughput in-line optical inspection at production speeds imposes tighter speed and accuracy constraints, which are not achievable by such conventional systems at a practical cost.
Although OFF has been considered to be a candidate to achieve both speed and detection accuracy concurrently, its application has not been considered practical due to a range of problems. As the size and contrast of the target defects to be detected diminishes, the accuracy, optical quality, and contrast of the masking pattern used as the optical filter becomes crucial. Achieving this type of quality in the past has been impossible except through the use of either static masks (such as photographic films or holographic gratings) or through static or difficult-to-configure mechanical systems. Static masks remain unsuitable for the presently considered application domain, since the pattern on the material to be inspected often needs to be changed during the regular operation of an AOI instrument and static masks cannot be quickly reconfigured to inspect the object with the new pattern or compensate for the varying orientation of the same pattern.
Re-configurable mechanical systems such as the ones described in U.S. Pat. No. 5,742,422 to Drake and U.S. Pat. No. 5,970,168 to Montesanto et al. attempt to partially solve this problem while preserving optical quality. However, the resulting systems are unacceptably slow to reconfigure. Moreover, they are constrained with respect to the realizable masking patterns, usually limited to a restricted number of horizontal and vertical lines. The use of lines as the masking pattern also causes its own adverse effects, manifesting as parasitic ghosts of defects in the filtered images. Furthermore, the mechanical implementations of the Fourier filters are complex, costly, unreliable, and unacceptably large in size and, therefore, cannot be easily expanded. For example, it is desirable to provide for multiple parallel inspections of a large surface area within tight tact time requirements.
Electrically programmable SLM devices first emerged in the last decade, at first showing some promise in solving the high speed filter reconfiguration problem. However, the available devices, such as light valves and early forms of transmissive LCDs, could not meet desired performance specifications. These requirements include:                High contrast ratio and high optical quality (transparency/reflectivity and optical uniformity), to achieve very high spectral suppression and undistorted signal transmission, leading to desired detection limits;        High spatial resolution and small size, to be feasible within compact optical arrangements and thus make multiple channel operation realizable;        A large fill-factor to minimize mask surface effects (having the structure of a grid) which causes undesired parasitic diffraction patterns on the filtered image; and        Operability within a large range of incidence angles, required by large Field of View (FOV) optical channel arrangements, enabling the coverage of a larger inspection surface area with a smaller number of inspection channels.        
Therefore, these devices were not practical for use in the high-throughput, in-line optical inspection of large flat patterned media at production speeds, while also satisfying industry requirements for detection limits. Thus, the promise was unfulfilled.